Related Experiment Videos
A comparative evaluation of methods of adjusting GPA for differences in grade assignment practices
Pui-Wa Lei1, Dina Bassiri, E Matthew Schulz
1Department of Educational and School Psychology and Special Education, 201 CEDAR Building, The Pennsylvania State University, University Park, PA 16802, USA. puiwa@psu.edu
Journal of Applied Measurement
|April 18, 2003
Summary
This study found that adjusted grade point average (adjusted-GPA) models, particularly those using Item Response Theory (IRT), offer better prediction from pre-admissions data than raw GPA. Adjusted-GPA provides more consistent student rankings within courses.
Area of Science:
- Educational Measurement
- Psychometrics
- Higher Education Research
Background:
- Traditional Grade Point Average (GPA) does not account for varying grading standards across academic institutions.
- Adjusted GPA (adjusted-GPA) aims to provide a more equitable measure by controlling for differences in course and departmental grading scales.
- Previous research suggests adjusted-GPA is more predictable from pre-admissions variables and offers more consistent student rankings.
Purpose of the Study:
- To compare the effectiveness of four polytomous Item Response Theory (IRT) models and three linear models in constructing adjusted-GPA.
- To assess the cross-validated performance of these adjusted-GPA models, including regression weights and course parameter estimates.
- To evaluate the predictive validity and consistency of adjusted-GPA measures against traditional raw GPA.
Main Methods:
- Employed four polytomous Item Response Theory (IRT) models: graded response model, rating scale model, partial credit model, and a two-parameter logistic model.
- Utilized three linear models for comparison in constructing adjusted-GPA.
- Conducted cross-validation of regression weights and course parameter estimates to ensure generalizability of the adjusted-GPA models.
Main Results:
- Adjusted-GPA demonstrated significant advantages over raw GPA in cross-validation, particularly in predicting pre-admissions variables.
- The highest predictive advantages were observed when adjusted-GPA was derived using the rating scale and partial credit IRT models.
- The graded response IRT model yielded the weakest cross-validity for adjusted-GPA construction.
Conclusions:
- Item Response Theory models, specifically the rating scale and partial credit models, are effective for creating adjusted-GPA measures with superior predictive validity.
- Adjusted-GPA offers a more robust and equitable academic performance metric compared to raw GPA, especially when accounting for grading leniency.
- The choice of IRT model significantly impacts the cross-validity and predictive power of the resulting adjusted-GPA.